| Summary: | GCC incorrectly rounds hex floats | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Neil Booth <neil> |
| Component: | middle-end | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | RESOLVED FIXED | ||
| Severity: | minor | CC: | gcc-bugs, jsm28 |
| Priority: | P3 | Keywords: | diagnostic, wrong-code |
| Version: | 4.1.0 | ||
| Target Milestone: | 4.1.0 | ||
| Host: | Target: | ||
| Build: | Known to work: | ||
| Known to fail: | Last reconfirmed: | ||
| Bug Depends on: | |||
| Bug Blocks: | 16989 | ||
Subject: Bug 21720 CVSROOT: /cvs/gcc Module name: gcc Changes by: jsm28@gcc.gnu.org 2005-07-29 01:24:39 Modified files: gcc : ChangeLog real.c gcc/testsuite : ChangeLog Added files: gcc/testsuite/gcc.dg: hex-round-1.c Log message: PR c/21720 * real.c (real_from_string): Set last bit if there is a nonzero hex digit beyond GCC's internal precision. testsuite: * gcc.dg/hex-round-1.c: New test. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.9593&r2=2.9594 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/real.c.diff?cvsroot=gcc&r1=1.155&r2=1.156 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5847&r2=1.5848 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/hex-round-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1 Fixed. It turns out my patch actually fixed a similar problem with the code handling digits before the ".", and my testcase didn't have enough zeroes to show the problem on 64-bit hosts. Testing a patch to fix the problem properly (by making the same change to the code for digits after the "." and expanding the testcase). Subject: Bug 21720 CVSROOT: /cvs/gcc Module name: gcc Changes by: jsm28@gcc.gnu.org 2005-07-29 21:14:22 Modified files: gcc : ChangeLog real.c gcc/testsuite : ChangeLog gcc/testsuite/gcc.dg: hex-round-1.c Added files: gcc/testsuite/gcc.dg: hex-round-2.c Log message: PR c/21720 * real.c (real_from_string): Also set last bit if there is a nonzero hex digit beyond GCC's internal precision after ".". testsuite: * gcc.dg/hex-round-1.c: Test more cases. * gcc.dg/hex-round-2.c: New test. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.9606&r2=2.9607 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/real.c.diff?cvsroot=gcc&r1=1.156&r2=1.157 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5852&r2=1.5853 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/hex-round-2.c.diff?cvsroot=gcc&r1=NONE&r2=1.1 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/hex-round-1.c.diff?cvsroot=gcc&r1=1.1&r2=1.2 Fixed. |
Unlike decimal floats, correctly rounding hex floats is easy and cheap, so we should do it. Moreover the standard requires it. With -Wall GCC complains that the following function is missing a return statement because it has incorrectly folded the comparison to false. int foo(void) { if (0x1.0000010000000000000000000000000000000001p0f != 1) return 1; } Remove one of the long line of zeroes and the precision falls within GCC's limits and it works.